CheckmateSample

This Python module provides a function to generate a checkerboard pattern using NumPy. The function make_checkerboard creates a 2D NumPy array representing a checkerboard with customizable board size and square (and rectangle) sizes. Can be used for geospatial ML problems to try and combat autocorrelation. The functions can export two (testing/training) and three (testing/training/validation) checkerboard patterns

Parameters

board_size (tuple[int, int]): Size of the board in (rows, columns).

square_size (tuple[int, int]): Size of each square in (rows, columns).

separation_size (int): Size of separation distance in pixels between squares.

validation (bool): If True, use a ternary (0, 1, 2) pattern; if False, use a binary (0, 1) pattern

an example of this is:

checkerboard_01 = make_checkerboard(board_size=(70,70), square_size=(10,10), separation_size=2)

xarray only:

dim_names (dict): Dictionary specifying the names of x and y dimensions da (xr.DataArray): Input DataArray to apply the checkerboard pattern to

an example of this:

ds = xr.tutorial.load_dataset('air_temperature')

air_temp = ds.air.isel(time=0)

square_size = (7, 5)  # 7x5 pixel squares
checkerboard_temp_0 = make_checkerboard_xr(air_temp, square_size, separation_size=3, keep_pattern=0, validation=False)

Dependencies

numpy and xarray

Useage

Create a NumPy checkerboard

checkerboard = make_checkerboard((10, 10), (2, 2))

Create an xarray DataArray

da = xr.DataArray(np.random.rand(100, 100), dims=['lat', 'lon'])

Apply checkerboard pattern to DataArray

result = make_checkerboard_xr(da, (10, 10), keep_pattern=1)